Discussion of Michael Ehrmann’s
“Targeting Inflation from Below:
How Do Inflation Expectations Behave?”
Eric T. Swanson
University of California, Irvine
Reflections on 25 Years of Inflation Targeting
Reserve Bank of New Zealand
December 1, 2014
Forward Nominal Interest Rates: UK vs. US
Forward 1-year Nominal Interest Rate, from 9 to 10 Years Ahead
10
United Kingdom
United States
9
percent
8
7
6
5
4
3
1992
1993
1994
1995
1996
source: Gurkaynak, Levin, and Swanson (2010 JEEA)
1997
1998
1999
2000
2001
2002
2003
2004
2005
Great Data: 3,000 Forecast Observations
Regress Forecast Dispersion on IT Indicator
But standard errors for some coefficients look low.
t-Statistics of 10, 20, and 40 Are Too High
Comment 1: Correct for Residual Autocorrelation
Use a panel Cochrane-Orcutt procedure:
Estimate serial correlation:
𝜀𝑐,𝑡 = 𝜌𝑐 𝜀𝑐,𝑡−1 + ν𝑡
Then use estimated 𝜌𝑐 to transform data:
Ω𝑐,𝑡 − 𝜌𝑐 Ω𝑐,𝑡−1 = 1 − 𝜌𝑐 𝛼𝑐 + 1 − 𝜌𝑐 𝛼𝑚
+ 𝛾1 𝐸𝑐,𝑡 𝜋𝑐,𝑡+ℎ1 − 𝜌𝑐 𝐸𝑐,𝑡−1 𝜋𝑐,𝑡+ℎ1−1
+ 𝛾2 𝐼𝑇𝑐,𝑡 − 𝜌𝑐 𝐼𝑇𝑐,𝑡−1 + 𝜈𝑡
Coefficient estimates shouldn’t change much, and standard
errors will be correct (but larger).
Comment 1: Correct for Residual Autocorrelation
But with larger standard errors, some estimates will no longer
be statistically significant:
Side Comment: Think about Month Fixed Effects
For forcast dispersion, month fixed effects are appropriate:
e.g., in November 2014, forecasters should largely agree
about 2014 inflation; in January 2014, not so much.
But for h-year-ahead forecasts, month effects make little sense:
e.g., no reason to think the average forecast of inflation in
2019 should be higher or lower in November than in January.
Downward Time Trends
Forecast dispersion may have downward trend over time:
• Greater central bank credibility
• Greater central bank transparency
– U.S. Fed is an example (Swanson, 2006 JMCB)
• Private sector learning about monetary policy
• Better private sector forecasting technology
• Great Moderation
Inflation Targeting observations occur later in the sample
Hard to disentangle effects of IT from these other factors
Most convincing way: some kind of control variable
GDP as a Potential Control Variable
GDP seems like a potential control variable:
• effects of IT on GDP volatility go in both directions
• net effect is probably smaller than effect of IT on inflation
But Michael’s results are 3X stronger for GDP than for inflation
Comment 2: Try to Find a Control Variable
Is there a control variable that makes the case that estimated
effects are due to IT?
GDP is one such potential control variable. But it suggests
that trends in other factors are more important than IT.
(Hard to argue that the effects of IT on GDP should be larger
than the effects of IT on inflation.)
Michael’s empirical results would be strengthened a lot if there
was a good control variable.
Central Bank Responsiveness to Inflation
Paper asserts that IT central banks need to respond less to
inflation:
• “if the central bank threatens to be more aggressive on inflation, it will
have to move rates by less in equilibrium.” (p. 3)
• “with inflation expectations anchored at target, policy rates need to react
less to changes in inflation.” (p. 22)
Theoretically, this assertion is suspect:
𝑟𝑡 = 𝑖𝑡 − 𝐸𝑡 𝜋𝑡+1
If a shock causes 𝐸𝑡 𝜋𝑡+1 to increase when policymakers want
lower 𝑟𝑡 , then 𝑖𝑡 under IT needs to respond more, not less
• e.g., some supply shocks
– “increases in oil prices today are more likely to promote consideration
of increased policy ease” (Bernanke, 2003 speech)
Central Bank Responsiveness to Inflation
Empirically, the assertion is also suspect:
Comment 3: Central Bank Responsiveness
The idea that IT central banks need to respond less to inflation
seems to come from “strict inflation targeting” intuition.
But this intuition may not hold for flexible inflation targeters,
which is the relevant case in practice:
“For quite a few years now, however, strict inflation targeting has been
without significant practical relevance. In particular, I am not aware of any
real-world central bank (the language of its mandate notwithstanding) that
does not treat the stabilization of employment and output as an important
policy objective.”
(Bernanke, 2003 speech)
The paper should soften (or at least justify better) its claims
about monetary policy responsiveness to inflation.
A Few Side Comments
Definition of IT is often subjective:
• UK adopted IT in 1992, but BoE not independent until 1997-8
• Not clear whether or when US should be classified as IT
Adoption of IT is endogenous:
• Creates sample selection problem
• If high-inflation countries adopt IT, and there is mean
reversion, then effects of IT will be overestimated (Ball and
Sheridan, 2005)
Today, IT is arguably just a method of communicating optimal
monetary policy.
Difference between IT and non-IT central banks is arguably
just one of communication, not substance.
Summary
Main Comments:
1. Correct standard errors for residual autocorrelation.
2. Try to find a control variable to help distinguish effects of IT
from other factors that improved over time.
3. Either soften or justify claim that IT implies less central bank
responsiveness to inflation.
Minor Comments:
• Think about and clean up month fixed effects
• The IT classification is subjective: check for robustness
• Discuss potential problems of IT endogeneity, mean
reversion
• Is the IT vs. non-IT distinction meaningful? Or is it just a
difference in transparency and communication?